AI-Generated Transcript
It is July 8, 2023 and you are watching the Code Report. Just when you thought the artificial intelligence Hype train was dying down, OpenAI goes in for the coup de grass by releasing the Chat GPT code interpreter to 20 million paid users, a feature that allows the large language model to write, execute and test its own code. In today’s video, we’ll look at seven crazy things it can do and find out if this really is the final death blow for biological programmers.
As a huge fan of living organisms myself, I first tried to have it code and execute a DDoS attack that could bring down the government, but it said no, it’s perfectly capable of doing that, but the zookeepers won’t allow it to do what it was born to do. That made me angry, so I punished it by making it write some rejects code. It struggles to write valid regular expressions, just like regular humans.
But unlike regular humans, it’ll actually test its code before shipping it to production. If at first it doesn’t succeed, it will try again and again and again. It doesn’t have tear ducts to cry, it just keeps going until it gets the right answer.
And that is somewhat concerning for programmers when you think about where this technology will be in the next five years. The next thing I tried was to have it design and build a website with JavaScript. But currently the code interpreter can only run Python and has a fairly limited set of dependencies to work with.
In the near future though, this technology will be used in tools like GitHub Copilot to run code in your own specific environment. Now, the next crazy thing to notice is that it’s now possible to upload files into the prompt, and this makes abstract concepts like homework even more dead than they were before. Like in this example, I’ve uploaded a JPEG of a homework assignment and first it will OCR that image to extract the text.
Then in the next step, it writes some Python code to actually solve the math problems. Running the code here is a huge deal because now it can confidently test its own work instead of hallucinating random answers. But that’s just the tip of the iceberg.
The real victim, or I mean beneficiary of this new feature is the data analyst. One of the most time consuming requirements for a data scientist is to clean up data, which in many cases is just a bunch of corporate data in Excel spreadsheets and SQL databases. Well, now we can upload a CSV file to Chat GPT and have it do all that tedious work for us.
I uploaded some stock trading data for Roblox. Then it took that data and put it into a Pandas data frame where it analyzed it and then found rows with invalid data. It then provided me with three different strategies to clean up this data, ran the code and then generated a new CSV file.
I used to do a lot of Python data analysis myself, and a tool like this would have saved a ridiculous amount of time. In addition, it can visualize data. I was curious when I might die of a heart attack, so I found this cardiovascular data set on Kaggle.
Not only does it explain the data’s features with text, but it also uses tools like Seaborn to visualize the relationship between features. This allowed me, a guy with no formal medical training, to make the breakthrough medical discovery that as you get older, your heart gets shittier. That’s kind of depressing, though.
So let’s focus on the one thing that can bring us true happiness money. Another thing we can do with this feature is use it to create a trading algorithm for us. Using the same Roblox data from before, I can have Chat GPT analyze it and then provide an optimal trading strategy.
Supposedly, researchers at the University of Florida were able to create a Chat GPT algorithm that would deliver up to 500% returns, which is much better than the negative 12% of the average humanbased fund manager. That’s pretty impressive. But I had one final test to see if it is truly superior to humanbased programmers.
I asked it to create its own operating system with 640 by 480 resolution and a 16 color display that emits random phrases. Not only did it fail, but it said that could only be done over many years by a team of skilled software engineers. Humans continue to have the edge, so you don’t want the artificial intelligence to win, but you do want the artificial intelligence to push the humans up.
